AI video · checked 2026-10-13

Can I vibecode HeyGen?

NOT REALLY· don't bother

HeyGen sells its avatar, voice and translation models, the GPUs that run them, and a library of licensed stock presenters. Open models can now lip-sync a recorded clip of you to a cloned voice, and an agent can wire that into a local pipeline over a weekend. You need an NVIDIA GPU, the result looks stiffer, and every minute takes a while to render. For occasional clips, HeyGen's free tier or a rented GPU beats building a studio at home.

$29/moprice

$348it costs you a year

weekendbuild time

—people replaced it

Before you start

What you need on hand.

DifficultyAdvanced · Real infrastructure or OAuth work.

TimeA weekend, mostly drivers, CUDA and model downloads

Claude Code, Codex or CursorAny of them can run the prompt.
A Windows or Linux PC with an NVIDIA GPULatentSync 1.5 needs at least 8 GB of VRAM. Macs can't run it locally; rent a GPU instead.
Python 3.11, ffmpeg and several GB of diskFor the voice, lip-sync and transcription models.
A 10-30 second clip of yourself and a voice sampleEven light, facing the camera, mouth visible, little background noise.

Before you buildFor a few short clips a month, HeyGen Free (3 videos up to a minute each) may be all you need.

1. Install one agent: Claude Code, Codex, Cursor.
2. Make an empty folder and open it in the agent.
3. Paste the prompt below, press enter, and answer its questions as it builds.

The prompt

Build me a local talking-head video maker to replace HeyGen for simple presenter
clips of myself. Requirements:

- Python 3.11 with a Gradio page on http://127.0.0.1:7860, and the same steps as a CLI.
- Inputs: a 10-30 second clip of me facing the camera (the base footage), a 10-20
  second sample of my voice, and a script typed into the page.
- Voice: generate the narration with Chatterbox TTS cloned from my sample. Split
  long scripts at sentence ends and join the pieces with ffmpeg.
- Lip sync: run LatentSync 1.5 on the base clip with the new audio, looping the
  clip forwards and backwards when the audio is longer. It needs an NVIDIA GPU with
  at least 8 GB of VRAM; if there is none, say so and stop instead of running on CPU.
- Captions: transcribe the result with faster-whisper, write an .srt, and
  optionally burn the captions into the video with ffmpeg.
- Save each render to ./renders/YYYY-MM-DD-<slug>/ with video.mp4, voice.wav,
  captions.srt and script.txt.
- Only my own face and voice: the page asks me to confirm that before cloning.
- No accounts, no telemetry, local-first; models download once from Hugging Face.
- Out of scope: stock avatars, photo-only avatars, live streaming avatars and
  translated dubbing.
- README: GPU driver and CUDA setup, how long a one-minute render took on my GPU,
  and an honest note that head and body motion will look stiffer than HeyGen's.

Opens your agent with the prompt filled in, you press enter. Or copy it and paste it into an empty folder.

Share on X

Max MiskowSponsoredAI agents, automations and MVPs, built for your company.Visit site ›

Then keep going

Next prompts, same folder.

The first prompt builds the core. Run these one at a time in the same folder to grow it into your own HeyGen.

Add batch rendering to the talking-head video maker in this folder. Requirements:

- Take a CSV and a script template with {{column}} fields, such as
  "Hi {{first_name}}, I saw {{company}}...".
- Render one video per row into ./renders/batch-YYYY-MM-DD/<row-number>-<slug>/.
- Reuse the cloned voice and base clip; never re-download models per row.
- Show progress, keep going when a row fails, and write a report CSV with the
  output path or error for each row.
- Update the README.

Add slide videos to the talking-head video maker in this folder. Requirements:

- Take a PDF of slides and a script with a --- separator between slides.
- Render each slide to a 1920x1080 image with pdftoppm, render the narration and
  lip-synced clip per section, and show the slide full-screen with me in a rounded
  corner box using ffmpeg overlay filters.
- Cut to the next slide when its section of narration starts.
- Captions as before, placed above my picture.
- Update the README.

Make the talking-head video maker in this folder run its GPU steps on Modal. Requirements:

- Keep the Gradio page local; move the Chatterbox and LatentSync steps into a Modal
  function with the models cached in a Modal volume.
- Read the Modal token from the CLI login, and GPU_TYPE from .env.
- Upload the inputs, download video.mp4 and the audio back into ./renders/ as before.
- Print the GPU seconds used and the estimated cost after each render.
- Update the README with Modal setup and an honest note on cold-start time.

What you lose

Natural head and hand movement, not just moving lips
Hundreds of stock avatars and avatars made from a photo
Video translation and voices in 30+ languages
Fast cloud renders without a GPU of your own
Templates, brand kits and team review

Why people still pay

Marketing, training and sales teams pay because one person can produce a presenter video in many languages without a camera, a studio or an actor. The output is good enough for courses and outreach, and the consent step for custom avatars makes it easier to use real faces at work.

Moatproprietary models infrastructure scale content & rights

Do the math

HeyGen or your own copy?

Over

Your usage

Count my time12h ×$/h

HeyGen, 1 yr$348

Your own copy, 1 yr$190running + build session

You keep$158pays for itself in the first month

Estimate · Free on a GPU you already own, apart from electricity. Without one, a rented cloud GPU bills by the hour, which adds up with long or frequent videos. The build session is token usage on your agent plan.

Leaving HeyGen

Take your data, then cancel.

1 · Take your data out of HeyGen

1Open Projects, click the ... button on a video and choose Download.
2Inside a video, the button in the top-right corner also downloads just the audio or the captions as an SRT file.
3Choose 720p, 1080p or 4K under Advance Settings; 4K needs a plan that allows it, so download full-quality copies while you still pay.
4There is no bulk download, so save each video you want to keep one by one.

2 · Cancel HeyGen

1Click your profile picture in the bottom-left corner and open Settings > Plan & Billing.
2Click the arrow next to Manage Billing and choose Cancel Plan (or Pause Plan), then finish the short survey.
3Subscribed in the iOS or Android app: cancel in App Store or Google Play subscriptions instead.

Access runs to the end of the cycle; unused credits expire then, and there's no prorated refund, including on annual plans.
Pausing moves you to Free for about 30 days, wipes the credit balance and bills again automatically afterwards.
On Free you keep one avatar slot; the others are locked, not deleted.
Cancelling a legacy unlimited plan is permanent: coming back means a current credit-based plan.
Deleting the account is a separate step under Settings > Account.

Checked 2026-10-13 against HeyGen's docs: 1, 2, 3

The escape hatch

Don't feel like building it? These are already free.

Duix.AvatarFreeOffline toolkit (formerly HeyGem) that clones your look and voice from a recording, then drives the clone with text or audio; needs an NVIDIA GPU. Community license.★ 16k · active Apr 2026 · docker to self-host
Windows, LinuxVisit ↗LivePortraitOpen sourceAnimates a still portrait from a driving video of a face; useful for photo avatars, not driven by audio on its own.★ 19k · active Jun 2026 · self-host, real ops
Windows, Linux, macOSVisit ↗InfiniteTalkOpen sourceAudio-driven talking video of any length from an image or a video; needs a large GPU.★ 8.1k · active May 2026 · self-host, real ops
LinuxVisit ↗

All 6 free alternatives to HeyGen ›

The numbers

HeyGen pricing

Plan	Monthly	Annual, per month	What you get
Free	$0	$0	3 videos a month up to 1 minute each, 1 custom avatar, up to 3 photo avatars, 1 voice clone, 1 seat.
Creator	$29	$24	600 credits a month, videos up to 30 minutes, unlimited photo avatars, no watermark.
Pro	$49	—	From 1,000 credits a month (larger credit tiers available), 4K export, videos up to 30 minutes.
Business	$149/workspace	—	1,500 credits a month, videos up to 60 minutes, 5+ custom avatars.
Enterprise	custom	—	No maximum video length, unlimited voice cloning.

Free tier: 3 videos a month, 1 minute each, 1 custom avatar, 1 seat.
Billing: Monthly or annual. Monthly credits roll over for one month; annual credits build up until renewal.
Hidden costs: How many credits a video uses depends on the model, its length and complexity, and the page gives no per-minute rate. Unused credits are lost when you cancel, with no prorated refund on monthly or annual plans. The API is priced separately.
Source: www.heygen.com

Also on the list

ElevenLabs$6/mo · AI audioKINDA

Descript$24/mo · Audio & videoKINDA

Loom$18/mo · Screen recordingKINDA

Questions

HeyGen, briefly

Not answered here? Tell us and we'll add it.

Ask a question ›

Free makes 3 videos a month, each up to 1 minute, with one custom avatar.

Not really. Marketing, training and sales teams pay because one person can produce a presenter video in many languages without a camera, a studio or an actor. The output is good enough for courses and outreach, and the consent step for custom avatars makes it easier to use real faces at work. The prompt on this page builds the closest honest substitute.

About $29 a month on the Creator plan, $348 a year (checked 2026-10-13).

Natural head and hand movement, not just moving lips; Hundreds of stock avatars and avatars made from a photo; Video translation and voices in 30+ languages; Fast cloud renders without a GPU of your own; Templates, brand kits and team review. If any of those matter to you, keep paying.